SSD isn't mind blowingly fast, Actually slower than 3 x Raptor HDD Raid Array

[Ketchup]

Several experts have also posted on a thread I made in another forum on this issue, and all agree with me as well that IOPS does not equal throughput.
http://forums.2cpu.com/showthread.php?t=99274

So you start a thread on another site about how wrong people on this site are. Then you link us to it.

And you continue to post here, why?

Hint: that called RUDE.

 

[Cerb]

Cerb: As I said several times already, seek times are at the heart of this matter (as is NCQ), as you have just shown and said. Thank you for proving my point for me. SSDs are better at IOPS than hard drives are.

If seek times are at the heart of it, then seek times are. Seek times can be high, even with high IOPS, though, and low with low IOPS.

It hasn't proven anything of yours correct. IOPS are one measure of throughput, just like MB/s is. For a given workload, MB/s is equivalent to some IOPS rating, and we use that.

IOPS are just another measure of throughput. A formula to show exactly that has been put forth in this thread multiple times. Sequential transfers are not necessary for it to be true, simply transfers of agreed-upon size and layout (IE, 64KB, random access).

Also, you've now changed the goalposts. That SSDs can sustain more IOPS than HDDs was not the matter in question, but the statement that IOPS and throughput were not equivalent measures of performance. Nobody argues that HDDs can sustain more IOPS than SSDs, nor has argued that.

For example, if IOPS and throughput are different, should we start considering CrystalDiskMark screenshots useless, as diagnostic aids? After all, it gives results in a throughput measurement, not IOPS. Of course not. The throughput results tell the same story, just with a different dimensionality.

 

[jaqie]

So you start a thread on another site about how wrong people on this site are. Then you link us to it.

Hint: I started that far before anyone else chimed in. Check the post time(s). Also you need to check the thread - I spoke of nobody except the anandtech SSD editor, calling his professional knowledge into question, nobody else's. That's called common sense.

 

[jaqie]

If seek times are at the heart of it, then seek times are. Seek times can be high, even with high IOPS, though, and low with low IOPS.

It hasn't proven anything of yours correct. IOPS are one measure of throughput, just like MB/s is. For a given workload, MB/s is equivalent to some IOPS rating, and we use that.

IOPS are just another measure of throughput. A formula to show exactly that has been put forth in this thread multiple times. Sequential transfers are not necessary for it to be true, simply transfers of agreed-upon size and layout (IE, 64KB, random access).

Also, you've now changed the goalposts. That SSDs can sustain more IOPS than HDDs was not the matter in question, but the statement that IOPS and throughput were not equivalent measures of performance. Nobody argues that HDDs can sustain more IOPS than SSDs, nor has argued that.

For example, if IOPS and throughput are different, should we start considering CrystalDiskMark screenshots useless, as diagnostic aids? After all, it gives results in a throughput measurement, not IOPS? Of course not. The throughput results tell the same story, just with a different dimensionality.

So because torque and horsepower are two different but related metrics in car engines, we should call their ratio the difference and merge torque and horsepower into the same measurement? There is a very good reason that IOPS and throughput are separate metrics. I never changed the goalposts, this is what started the whole thing, check my original statement which I have been trying to explain time and again to people that don't understand it. You are trying to change the goalposts. I never have, I have stayed true to my original intent and statement from my first post in this thread.

---------------------------------------------------------------------------
This is beginning to devolve into a mud slinging fest, and this is the point I am leaving the discussion. Have fun, kids.

I agree. But to be frank jaqie, we're of the opinion that you're the one who has egged it on. Everyone - including you - knows the rules and is expected to be on their best behavior.
-ViRGE

 

[Cerb]

So because torque and horsepower are two different but related metrics in car engines, we should call their ratio the difference and merge torque and horsepower into the same measurement?

No. That analogy is note remotely applicable.

A good analogy would be more like comparing the time it took to go from Atlanta to Los Angeles, including all the stops along the way, and your average miles per hour for the trip.

Miles per hour for the trip would be analogous to MB/s.
Stops per hour for the trip would be analogous to IOPS.

Make it 10 people going from other destinations to LA, and the analogy could even include command queuing.

For example, performing a 4K random read test, if an HD performs at 0.20MB/s, I can tell you that it is performing at a speed of 51 IOPS. If an SSD performs at 20MB/s, I could just as well tell you that it is performing at a rate of 5120 IOPS. Why? Because they have a fixed relationship, in any well-made test (which controls for data size and arrangement).

And, with overlapping IO (IE, NCQ), any measure over a significant amount of time (including a full second) effectively decouples seek time from the measure, making it its own separate issue to deal with (it is not a non-issue with SSDs, either, especially at high QDs--it may sometimes be best to reduce the QD limit to keep them very low).

 

[cheez]

ryderOCZ,

Under IDE ATA/ATAPI controllers in device manager tells the following:

Intel(R) ICH8R/DH/DO SATA AHCI Controller

Is there anything else you need? Should I go to Intel site and look for the latest driver there? Please help. I would like to have my missing ~20 MB/s in write score back.

cheez

ryderOCZ,

Did you get my response above? What can I do to get my ~20 MB/s sequential write speed? Is it because I have slow running CPU and FSB? I know that they have some impact on this benchmark from what I heard too. Please give me instruction what to do next.

Thanks a valleyball court,


cheez

 

[Makaveli]

ryderOCZ,

Did you get my response above? What can I do to get my ~20 MB/s sequential write speed? Is it because I have slow running CPU and FSB? I know that they have some impact on this benchmark from what I heard too. Please give me instruction what to do next.

Thanks a valleyball court,


cheez

Question how much space have you used on the drive currently?

Not sure if it was posted before.

 

[cheez]

Question how much space have you used on the drive currently?

Not sure if it was posted before.

Well, the day I ran the benchmark last time I had used up less than 10% of drive space. Right now at around 13%?..

I also had minimal number of background apps running, pretty much only the services needed for Windows OS, ATi Control Center, and sound card. Nothing fancy. CPU utilization at 0 to 1%.

I do have superfetch disabled. I didn't make any change in disk defragmenter service.

Windows 7 Pro 64bit with SP1.


cheez

 

[ryderOCZ]

ryderOCZ,

Under IDE ATA/ATAPI controllers in device manager tells the following:

Intel(R) ICH8R/DH/DO SATA AHCI Controller

Is there anything else you need? Should I go to Intel site and look for the latest driver there? Please help. I would like to have my missing ~20 MB/s in write score back.

We may not be able to get it back without a secure erase. What is the date of your drivers? Right click the line above (in device manager) and select driver and see the date it was published.
You do have a fairly old chipset and even with newer drivers, you may not get the speed. Newer drivers don't optimize older chipsets, they add functions to newer chipsets.

 

[cheez]

We may not be able to get it back without a secure erase. What is the date of your drivers? Right click the line above (in device manager) and select driver and see the date it was published.
You do have a fairly old chipset and even with newer drivers, you may not get the speed. Newer drivers don't optimize older chipsets, they add functions to newer chipsets.

I am at work right now but I remember the year for the date of the driver published. It's 2009.


So the scores you referenced in your link is with the newer chipset then? Yeah my PC is pretty dang old. It doesn't break on me and wanting to keep going. ^_^ That motherboard and CPU is freakin tank! They are 7 years old.

Are you suggesting secure erase will help? I don't mind doing that. But I would like to know if this will surely fix it... it's a quite a work getting OS loaded and configured again.

P.s. I have never done secure erase on this SSD.


cheez

 

[ryderOCZ]

I am at work right now but I remember the year for the date of the driver published. It's 2009.


So the scores you referenced in your link is with the newer chipset then? Yeah my PC is pretty dang old. It doesn't break on me and wanting to keep going. ^_^ That motherboard and CPU is freakin tank! They are 7 years old.

Are you suggesting secure erase will help? I don't mind doing that. But I would like to know if this will surely fix it... it's a quite a work getting OS loaded and configured again.

P.s. I have never done secure erase on this SSD.


cheez

There really isn't anything to "fix" because 3 weeks or 6 months from now you might be down that 20MB/s again through your pattern of usage. Very few drives stay at the very top of their performance 100% of the time. A secure erase will reset everything to out of the box performance though.

 

[cheez]

Ok thanks ryderOCZ. I'll just stick with this then.


Another question, does SSD need a defrag? I've been wondering about this. Will defragging help when the files are scattered everywhere say several months down the road?


cheez

 

[GlacierFreeze]

No. Make sure it's disabled.

 

[Cerb]

Another question, does SSD need a defrag? I've been wondering about this. Will defragging help when the files are scattered everywhere say several months down the road?

Not only no, but this would also be a benefit to a newer Windows version. With a decent amount of HDD space left, Vista and 7, and I'm sure 8, don't cause nearly the fragmentation that prior Windows versions did in the first place (7 and newer will also ignore SSDs, so the service can be left on, for HDDs).

Defragging will cause much more wear on the SSD than regular use will.

 

[aigomorla]

i have been on a 6 array raptor until i retired them out in lue of SSD's.

However my Raid0 SSD's absolutely destorys them in loadup time.
I dont have a 6 raptor drive array anymore, however im still using a R0 raptor for purely cache / temp folder reasons.

Raptors R0 (oh god 4k took forever it said 20 min to finish!! i canceled it in the middle of 4k testing.. cuz u get the point)

I think the access time wasnt refreshed from my previous SSD run.
But you can see from seq. times how big of a difference they are.

VS.

ForceGT in R0.

my 4k sucks... :\

ICH10R Both on the same motherboard SATA 2.

 

[Hellhammer]

my 4k sucks... :\

No scaling benefits at queue depth of 1 (which is what AS-SSD uses for the 4K test)

 

[yefi]

Because they aren't related in a fixed way.

Their relationship is dependent on transfer size, but all things being equal, they are interchangeable.

Saying an SSD shines in IOPS but not throughput, makes the assumption that when you say IOPS you mean small transfer sizes, and when you say throughput you mean large transfer sizes. Yet throughput can just as well mean 4K as it can mean 1024K transfers.

 

[Zap]

Now you can have sound in SLI!

He can listen to different stuff in each ear. Take multitasking to the next level! OVER 9000!!!

The original raptor was using a 1 Single 32GB platter!

The new green drives are using what 500GB platters?

You're halfway there. AFAIK latest drives have 1TB platters.

...

I thought this thread was on its way to becoming a "bicker endlessly about SSDs" but then along comes aigomorla with his usual "oh yeah, I have one too, and mine is faster."

 

[aigomorla]

but then along comes aigomorla with his usual "oh yeah, I have one too, and mine is faster."

lolol

wasnt u responsible for my initial ssd migration?


LOL anyhow i figured people would want to see some viable numbers of both drives in R0 on the same ICH controller.


OP: i am addicted to SSD's on R0.
I am one of the nerds who u always wonder how the hell did he load up so fast to get first line at the chopper in battlefield maps person.

 

[Makaveli]

lolol

wasnt u responsible for my initial ssd migration?


LOL anyhow i figured people would want to see some viable numbers of both drives in R0 on the same ICH controller.


OP: i am addicted to SSD's on R0.
I am one of the nerds who u always wonder how the hell did he load up so fast to get first line at the chopper in battlefield maps person.

I can agree with you on that even tho I'm on Raided Sata 2 ssd's I usually get into the map first. I've been playing L4D2 a lot recently after taking a few months off and I can get in the server and kill about 10 zombies outside the safe room before anyone even moves.

I even had someone else ask me what my setup was as he had a crazy SSD setup similar to Adam K's and he just barely beat me into the server and he said usually I'm in here for a least a min before any else joins and you were right behind me.

 

[AdamK47]

Ran the test on my 4TB Hitachi 7200 RPM drive. Took a while to complete.

Very much different compared to the 4TB Vertex 4 array.

 

[cheez]

You guys are all doing wrong... The AS SSD benchmark is optimized for the SSD drives not hard drives. It's not designed for the hard drives. Apples and Oranges guys.

 

[Ao1]

Not wishing to fan any flames, but here is a simple explanation.

“Service time” measures how long it takes to process a specific request
“Queue time” represents the amount of time a request waits for service
“Response time” = service time + queue time

As Hellhammer has stated: IOPS = [Bytes per second] / [IO size in bytes]

From the AS SSD screen shots below it possible to check this formula:

4 K QD1
4K Read MB/s = 26.75 MB/s
4K Read IOPS = 6849 IOPS
Converting that to bytes:
28049408/ 4096 = 6848 IOPS

4K QD 64
4K Read MB/s = 364.79MB/s
4K Read IOPS = 93387 IOPS
Converting that to bytes:
382510039.04/ 4096 = 93386.24 IOPS

Seq Read Speed (16MB xfer)
Seq 16 MB Read MB/s = 515 MB/s
Seq 16 MB Read IOPS = 32.19 IOPS
Converting that to bytes:
540016640/ 16777216 = 32.18 IOPS

Hellhammer’s formula seems to be working out quite nicely J



Uploaded with ImageShack.us



Uploaded with ImageShack.us

 

[Ao1]

Cheez, you have the worst SSD for the workload you are applying. HDD’s are great at seq writes whilst the OCZ Sandforce drive is not as it can’t compress the data. If you are only comparing non compressible writes then sure your HDD array will be faster. If you had a good SSD that could provide 400MB/s+ for seq write speeds you would not have seen an issue.

 

[Zap]

wasnt u responsible for my initial ssd migration?

I fed the beast? Oh teh NOES~! D:

I've been playing L4D2 a lot recently after taking a few months off and I can get in the server and kill about 10 zombies outside the safe room before anyone even moves.

I remember when I got my first SSDs with games installed. Three Intel 40GB in RAID0 (well, cheaper Kingston models flashed with Intel firmware). I decided to find out what would happen in L4D2. First level in, I think I nearly made it to the safe room. My teammates were not amused.